question about spectral normalization

[Honlan]

Thanks a lot for your awesome release.

May I ask a question about the implementation of SpectralNormalization?

The call function of it is as follows, where the the line self.restore_weights() is commented out.

Why should this line be commented? After update_weights, is it necessary to restore the weights?

def call(self, inputs, training=None, mask=None):
        self.update_weights()
        output = self.layer(inputs)
        # self.restore_weights()  # Restore weights because of this formula "W = W - alpha * W_SN`"
        return output